Validation of an AI-based Biliopancreatic EUS Navigation System for Real-time Quality Improvement: A Prospective, Single-center, Randomized Controlled Trial

Not Applicable

• Conditions: Bile Duct Diseases; Endoscopic Ultrasonography; Pancreatic Disease; Artificial Intelligence
• Interventions: Other: AI-based biliopancreatic EUS navigation system

• Registration Number: NCT05457101
• Lead Sponsor: Renmin Hospital of Wuhan University

Brief Summary

Endoscopic ultrasonography (EUS) is a key procedure for diagnosing biliopancreatic diseases. However, the performance among EUS endoscopists varies greatly and leads to blind areas during operation, which impaired the health outcome of patients. We previously developed an artificial intelligence (AI) device that accurately identifies EUS standard stations and significantly reduces the difficulty of ultrasound image interpretation. In this study, we updated the device (named EUS-IREAD) and assessed its performance in improving the quality of EUS examination in a single-center randomized controlled trial.

Detailed Description

In recent years, endoscopic ultrasonography (EUS) has developed into a preferred imaging modality for the diagnosis of biliopancreatic diseases, especially small (\< 3 cm) pancreatic tumors and small (\< 4 mm) bile duct stones. Therefore, EUS is often chosen as the main tool for screening early biliopancreatic diseases among high-risk individuals. However, a plenty of studies have shown that the detection rate of biliopancreatic diseases under EUS varies from 70% to 93% among different endoscopists due to examination quality and operators differences, which suggest that there are missed diagnosis of lesions. The missed diagnosis of pancreatic cancer makes patients lose the opportunity of radical surgery, and the five-year survival rate is reduced to 7.2%; and the missed diagnosis of choledocholithiasis causes severe acute diseases such asacute cholangitis and acute pancreatitis; it has serious consequences on the prognosis and quality of life of patients. Therefore it is important to reduce the missed diagnosis of lesions while further expanding the application of EUS.

Ensuring the examination quality is a seminal prerequisite for discovering biliopancreatic lesions in EUS. There are two main reasons affecting the quality of biliopancreatic EUS examination: First, non-standard operation by endoscopists; excellent biliopancreatic EUS examinations require the continuity and integrity of the scan. According to the experience of the Japanese Society of Gastrointestinal Endoscopy and European and American experts, multi-station approach in biliopancreatic EUS has been established as the standard scanning procedure. And these standard stations include anatomical landmarks that can be used to locate the transducer and identify areas that are not scanned. The American Society for Gastrointestinal Endoscopy (ASGE) and the American Association for Gastrointestinal Endoscopy (ACG) Endoscopic Quality Working Group have also issued quality indicators that should be completed for EUS examination. But they are often not well followed because of a lack of supervision and availability of practical tools, and there are a large number of blind areas in current daily EUS scans. Secondly, it is difficult in understanding US images with gray and white texture. Even experienced endoscopists have some challenges in identifying anatomical structures in EUS images. Therefore, it is critical to develop a practical tool that can monitor the blind area of EUS examination in real time, reduce the difficulty of ultrasonographic interpretation, and standardize the quality of EUS examination.

Deep learning has been successfully applied to many areas of medicine. In the field of endoscopic ultrasonography, most researches are dedicated to the use of computer tools to assist in the diagnosis of lesions in static images, while rare work studied the role of deep learning in monitoring the blind area of EUS examinations and exploring assistance on real-time ultrasonographic interpretation. Previously, we have successfully developed and validated an EUS navigation system that can identify the standard stations of pancreas and bile duct EUS in real time. Although encouraging preliminary results have been published regarding the use of artificial intelligence in reducing the difficulty of EUS images, this system has not been validated in a real-world clinical setting, and it is unclear whether it can be successfully applied in clinical practice and improve the quality of EUS examination.

Therefore, in this study, we updated the EUS-intelligent and real-time endoscopy analytical device (named EUS-IREAD) based on the aforementioned biliopancreatic EUS station recognition models and further trained an anatomical landmark identification function to better locate the transducer position and diagnose biliopancreatic lesions. We then conducted a single-center randomized controlled trial to assess its adjunctive performance to EUS endoscopists in a clinical setting.

Study Timeline

• Status Verified: 2022-07
• Study Start: 2022-07-01
• Study First Submitted: 2022-07-09
• Study First Submitted QC: 2022-07-09
• Study First Posted: 2022-07-13
• Last Update Submitted: 2023-06-19
• Last Update Posted: 2023-06-22
• Primary Completion: 2023-07-01
• Study Completion: 2023-07-30

Recruitment & Eligibility

• Status: UNKNOWN
• Sex: All
• Target Recruitment: 264

Inclusion Criteria

1. Male or female aged 18 or above;
2. Patients able to give informed consent were eligible to participate.
3. Able and willing to comply with all study process.
4. history of previous biliopancreatic disease
5. Biliopancreatic lesions suspected due to clinical symptoms and/or radiological findings and/or laboratory findings
6. Patients at high risk of pancreatic cancer : Known genetic mutations associated with pancreatic cancer risk (BRCA2, BRCA1, PALB2, ATM, CDKNA/p16); Familial pancreatic ductal adenocarcinoma without known germline mutation; Peutz-Jeghers syndrome (STK11); Lynch syndrome (MLH1/MSH2/MSH6, EPCAM, PMS2); Familial adenomatous polyposis (APC). etc.

Exclusion Criteria

1. Has participated in other clinical trials, signed informed consent and was in the follow-up period of other clinical trials.
2. Has participated in clinical trials of the drug and is in the elution period of the experimental drug or control drug.
3. patients with absolute contraindications to EUS examination;
4. Drug or alcohol abuse or psychological disorder in the last 5 years.
5. Patients in pregnancy or lactation.
6. bleeding diathesis or thrombocytopenia
7. history of previous digestive surgery.
8. severe medical illness
9. upper GI tract obstruction
10. previous medical history of allergic reaction to anesthetics
11. anatomical abnormalities of the upper gastrointestinal tract due to advanced neoplasia
12. Researchers believe that the patient is not suitable to participate in the trial.

Study & Design

• Study Type: INTERVENTIONAL
• Study Design: PARALLEL

Arm && Interventions

Group	Intervention	Description
with AI-based biliopancreatic EUS navigation system	AI-based biliopancreatic EUS navigation system	The endoscopists in the experimental group will be assisted by EndoAngel, which can in real-time prompt standard stations and anatomical structures during EUS.

Primary Outcome Measures

Name	Time	Method
Missed scanning rate of standard stations in the experimental group and control group	twelve month	It was calculated by dividing the number of standard stations that is not scanned by the number of stations that should be scanned.

Secondary Outcome Measures

Name	Time	Method
Missed scanning rate of standard stations and anatomical landmarks for individual	twelve month	the Missed scanning rate of standard stations and anatomical landmarks of biliopancreatic endoscopic ultrasonography in different endoscopists in the EUS-IREAD assisted group and control groups
Operation time	twelve month	In addition to puncture, elastography, enhanced ultrasound and other observation of lesions or treatment, the time used to observe the biliopancreatic system
Missed scanning rate per standard station	twelve month	It was calculated by dividing the number of patients who are not scanned at a station by the total number of patients who should be scanned at the station
Missed scanning rate of anatomical landmarks in the experimental group and control groups	twelve month	It was calculated by dividing the number of anatomical landmarks that is not scanned by the number of anatomical landmarks that should be scanned
Missed scanning rate of anatomical landmarks in different standard stations	twelve month	It was calculated by dividing the number of anatomical landmarks that is not scanned under a station by the number of important anatomical landmarks that should be scanned under that station

Trial Locations

Locations (1)

Renmin Hospital of Wuhan University; 🇨🇳 Wuhan, Hubei, China